Unveiling evolving nationalistic discourses on social media: a cross-year analysis in pandemic

The COVID-19 pandemic profoundly impacted global public discourse, with social media platforms like Twitter becoming central spaces for expressing opinions and emotions. Existing research has explored various aspects of online discourse during the pandemic, including the prevalence of negative emotions, the shaping of country image, racism, and nationalism. However, a significant gap exists in understanding the dynamic evolution of online discourse and the interplay between quantitative and qualitative methods in analyzing it. This study addresses these gaps by examining the evolution of nationalistic discourses on Twitter over three years (2020-2022), employing a multifaceted approach that integrates advanced quantitative techniques with a novel qualitative framework. The study's importance lies in its contribution to a deeper understanding of how nationalist sentiments evolve in response to global crises and how these sentiments are mediated and amplified through social media.

The literature review explores the complex relationship between nationalism, the internet, and social media. It examines existing scholarship on nationalism, noting the ambiguity surrounding its definition and its frequent intertwining with concepts like racism, patriotism, and populism. The review highlights the role of elites in propagating nationalist narratives and the increasing importance of online spaces in shaping national identities. It also considers the impact of the internet as both a 're-embedding' and 'dis-embedding' technology, capable of reinforcing or disintegrating identities. The review discusses the rise of internet-driven nationalism and its contribution to societal division. Finally, the review examines previous studies on sentiment analysis and topic modeling applied to COVID-19 related tweets, highlighting the predominance of negative emotions in the early stages of the pandemic and the lack of studies examining the dynamic evolution of this discourse.

This study utilizes a mixed-methods approach, integrating quantitative and qualitative analyses of a dataset of approximately 2.65 million tweets related to China and COVID-19 from 2020 to 2022. **Quantitative Analysis:** * **Sentiment and Emotion Analysis:** The study employs a fine-tuned transformer model developed by CardiffNLP to analyze the sentiment (positive, neutral, negative) and emotions (anger, joy, sadness, optimism) expressed in the tweets. The model is trained on various datasets and is capable of processing tweets with diverse linguistic features, including URLs and emoticons. Descriptive statistics are calculated to show the average sentiment and emotion scores over time. * **Topic Modeling:** Bertopic, an advanced topic modeling technique, is used to identify latent topics within the tweet data. This method employs pre-trained transformer-based language models and a class-based variation of TF-IDF. The number of topics was set at 15 with 25 n-terms. Hierarchical clustering is used to visualize the relationships between the identified topics. * **Network Analysis:** Co-occurrence networks of high-frequency words are constructed and analyzed using Gephi software to visualize the connections between key concepts in the discourse. **Qualitative Analysis:** * **Evolving Discourse Framework Analysis:** A novel qualitative framework is developed, drawing on frame theory and existing research on nationalism. This framework categorizes the discourse into three frames: ‘feeling,’ ‘identity,’ and ‘action.’ The analysis examines the evolution of these frames over time and the relationships between them. High-frequency word analysis and manual coding of highly retweeted tweets are used to further explore the content within each frame. The study integrates these quantitative and qualitative findings to develop a comprehensive understanding of the evolution of nationalistic discourses on social media during the pandemic.

The key findings of the study can be summarized as follows: 1. **Evolutionary Trends in Sentiment and Emotion:** Negative sentiment was relatively stable in 2020 and 2021 but surged significantly in 2022, while positive sentiment showed a consistent upward trend. Influential tweets (those with high retweet and favorite counts) exhibited a stronger negative bias than average tweets. Anger decreased progressively from 2020 to 2022, while sadness showed a surge in 2022. Joy and optimism peaked in 2021 and declined in 2022. 2. **Evolving Topic Clusters:** Topic modeling revealed shifts in the dominant themes over time. In 2020, the discourse focused on nation-state identity and pandemic-related hostility. In 2021, the focus shifted to the consequences of actions (e.g., protests). By 2022, the discourse became more fragmented, with increased attention on specific cities and specific global pandemic prevention strategies. 3. **Dynamic Co-occurrence Networks:** Network analysis revealed an increasing graph density over time, suggesting a more diverse and fragmented discourse in 2022. The analysis also showed a shift in the connections between key concepts over the three years. 4. **Nationalism Discourse Framework:** The qualitative analysis using the Evolving Discourse Framework revealed the prominence of the "feeling" frame (insecurity, anxiety), a weakening "identity" frame (national identity becoming less distinct over time), and a relatively consistent "action" frame (programs, strategies). Manual coding of highly retweeted tweets revealed sadness as the dominant emotion across all three years, with anger fluctuating over time and optimism showing a limited presence.

The findings of this study address the research question by providing a nuanced understanding of the evolution of nationalistic discourses on social media during the COVID-19 pandemic. The study demonstrates the dynamic interplay between emotions, identity, and actions in shaping online nationalist narratives. The increasing graph density and fragmentation of the discourse in 2022 suggest a complex and multifaceted landscape of online nationalist sentiments, which is further illuminated by the shifting prominence of the three frames identified in the study. The results highlight the significance of considering both quantitative and qualitative methods to gain a comprehensive understanding of online public discourse during crises. The study contributes to the field by providing a novel framework for analyzing the evolution of nationalist sentiments in online spaces and by demonstrating the utility of mixed-methods approaches in this area of research.

This study provides a valuable contribution to understanding the evolving nature of nationalist discourses on social media during the COVID-19 pandemic. The mixed-methods approach employed offers a robust methodology for future research in this domain. The findings highlight the need for further research using more advanced NLP techniques, such as self-supervised learning and AIGC models, to enhance accuracy and address limitations in sentiment and emotion analysis. Future studies should also focus on developing more comprehensive mixed-methods frameworks that integrate computational methods with social science theories effectively.

The study's primary limitations lie in the reliance on unsupervised machine learning for sentiment and emotion analysis, which introduces potential errors, and the inherent limitations of applying a pre-established theoretical framework to empirical data. The use of Twitter data as the primary source may not fully capture the diversity of online discourse, and the focus on China-related tweets limits the generalizability of the findings to other contexts.